How to setup ComfyUI AI Image Generation on your AMD RDNA 4 GPU using ROCm 7.1.1 on Windows

With the release of ROCm 7.1.1, RDNA 4 users can finally enjoy the full performance of their GPUs for AI tasks on Windows.

Here's Our Full Guide on Enabling ComfyUI on AMD RDNA 4 GPUs Using ROCm 7.1.1 on Windows

Previously, they were limited to Linux for native ROCm support, or forced to use workarounds like ZLUDA or WSL, but now the architecture has full support for PyTorch through ROCm (TensorFlow, Jax, and ONNX are still only supported on Linux for RDNA4).

This marks a massive leap forward for the usability of Radeon's latest offerings when it comes to tasks such as image generation, because it removes the need for WSL (Windows Subsystem for Linux), drastically increasing usability and performance. Whereas it once took users up to a month to set up ComfyUI and other frameworks for local AI image generation on Radeon GPUs, the process is now impressively straightforward, as we'll see below.

Initial Configuration

To start with, download and install the 25.20.0.17 drivers to enable support for 7.1.1 PyTorch on Windows. This driver is compatible with RDNA 3, RDNA 3.5, and RDNA 4 graphics solutions, although curiously, AMD only lists the 7900XTX out of the RDNA 3 lineup as compatible.

After you've done that, go ahead and install Miniconda, which you can find here. You'll also need Git in order to clone the latest ComfyUI build. Once you have both of these, launch Anaconda Prompt and create a new conda environment. This will help us keep track of all the installed dependencies and keep the process as simple as possible.

Run the following to create an environment: conda create --name insertnamehere python=3.12. The reason for using Python 3.12 (and not the latest versions of Python) is that ROCm 7.1.1 wheels for PyTorch are only compatible with 3.12, so it's important that you specify this during environment creation.

Activate the environment by running. conda activate insertnamehere.

Installing PyTorch for ROCm 7.1.1

Next, we'll install PyTorch in this environment. You can find detailed instructions here, but the gist of it is to run the following commands in your conda environment:

pip install --no-cache-dir ^
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_core-0.1.dev0-py3-none-win_amd64.whl ^
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_devel-0.1.dev0-py3-none-win_amd64.whl ^
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_libraries_custom-0.1.dev0-py3-none-win_amd64.whl ^
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm-0.1.dev0.tar.gz

Next, run the following commands to install sub-dependencies of PyTorch:

pip install --no-cache-dir ^
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torch-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl ^
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torchaudio-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl ^
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torchvision-0.24.0+rocmsdk20251116-cp312-cp312-win_amd64.whl

If everything is installed properly, you should be able to run python -c "import torch; print(f'device name [0]:', torch.cuda.get_device_name(0))" And see your GPU listed as the output. I'm running a 9060XT 8GB, and as you can see, our environment is now GPU aware through ROCm and therefore has access to it.

Installing ComfyUI

The next step is to navigate to the directory you want to install ComfyUI in. For example, if you want to install it on your Desktop, use cd Desktop. Then, run

git clone https://github.com/comfyanonymous/ComfyUI.git

Navigate into the install folder for ComfyUI by running cd ComfyUI. Then, run pip install -r requirements.txt to download all the dependencies for ComfyUI.

And with that, we're ready to launch ComfyUI! Run the command python main.py to start it up. You can access the GUI by clicking on the web server link while holding down Ctrl.

Setting up models

Once you've opened the GUI, go to Templates on the left, and search for "SDXL Turbo". Download the model when prompted:

Then, move the model from your downloads folder to ComfyUI > Models > Checkpoints

Add your prompt to the box labelled CLIP text encode (prompt) - the upper one. Press run and voila! You've successfully generated an AI image completely locally on your RDNA 4 GPU!

Here's an image I generated, using this prompt: beautiful landscape scenery glass bottle with a galaxy inside cute fennec fox snow HDR sunset. Try it out and show off your masterpieces in the comments down below!

Follow Wccftech on Google to get more of our news coverage in your feeds.